Knowledge-transfer-based learning framework for airspace situation evaluation

1. An airspace situation evaluation framework for an airspace, the evaluation framework comprising a server and a memory module connected to the server via a local area or wide area network, wherein the airspace includes a target sector and multiple non-target sectors, the memory module stores sector situation (SS) samples and airspace operation data including aircraft trajectory data and airspace configuration data, and the evaluation framework evaluates the airspace situation by executing steps including:

(a) generating target samples from the airspace operation data, storing the target samples in the memory module, and learning knowledge from the target samples using multi-factor subset generation and multi-base evaluator construction to build target base evaluators, wherein the target samples are sector situation samples derived from the target sector;

(b) generating non-target samples from the airspace operation data, storing the non-target samples in the memory module, and learning knowledge in the non-target samples using multi-factor subset generation and multi-base evaluator construction, together with a sample transformation, to build non-target base evaluators, wherein the non-target samples are sector situation samples derived from one or more of the non-target sectors; and

(c) integrating the target base evaluators and the non-target base evaluators based on evaluation confidence analysis of the target base evaluators and the non-target base evaluators.

2. The airspace situation evaluation framework of claim 1 , wherein each of the target or non-target samples comprises values of a plurality of airspace situation factors, and a situation index which is a continuous or discrete value reflecting a situation corresponding to the values of the plurality of airspace situation factors.

3. The airspace situation evaluation framework of claim 1 , wherein in step (b) the sample transformation is used to transform the non-target samples into transformed non-target samples in order to eliminate discrepancies between the transformed target samples and the transformed non-target samples, so that the non-target base evaluators built from the transformed non-target samples have higher confidence on evaluating the target samples than those non-target base evaluators built without the sample transformation.

4. The airspace situation evaluation framework of claim 3 , wherein the sample transformation for the non-target samples is implemented through a kernel embedding method.

5. The airspace situation evaluation framework of claim 4 , wherein the kernel embedding method is implemented as solving a transformation matrix optimization problem with optimization goals of: within a common sample space into which the target and non-target samples are transformed, (i) minimizing discrepancies between the transformed target samples and the transformed non-target samples of a same category; and (ii) preserving the volume of knowledge for SS evaluation contained in the transformed non-target and target samples, which is measured by a dimension-category correlation degree of the transformed target and non-target samples.

6. The airspace situation evaluation framework of claim 1 , wherein in step (c) a weight of each of the base evaluators on evaluating a target sample is assigned as the confidence of the base evaluator for evaluating the target sample, then selecting out a portion of the base evaluators with higher weights among all the target and non-target base evaluators.

7. The airspace situation evaluation framework of claim 6 , wherein the weights of the base evaluators are calculated with respect to each of the target samples to be evaluated.

8. A method of evaluating an airspace situation for an airspace having a target sector and multiple non-target sectors, the method comprising the steps of:

(A) providing a server and a memory module connected to the server via a local area or wide area network, wherein the memory module stores sector situation (SS) samples and airspace operation data including aircraft trajectory data and airspace configuration data;

(B) the server generating target samples from the airspace operation data, storing the target samples in the memory module, and executing a software program to learn knowledge from the target samples using multi-factor subset generation and multi-base evaluator construction to build target base evaluators, wherein the target samples are sector situation samples derived from the target sector;

(C) the server generating non-target samples from the airspace operation data, storing the non-target samples in the memory module, and executing the software program to learn knowledge in the non-target samples using multi-factor subset generation and multi-base evaluator construction, together with a sample transformation, to build non-target base evaluators, wherein the non-target samples are sector situation samples derived from one or more of the non-target sectors; and

(D) the server integrating the target base evaluators and the non-target base evaluators based on evaluation confidence analysis of the target base evaluators and the non-target base evaluators.

9. The method of claim 8 , wherein each of the target or non-target samples comprises values of a plurality of airspace situation factors, and a situation index which is a continuous or discrete value reflecting a situation corresponding to the values of the plurality of airspace situation factors.

10. The method of claim 8 , wherein in step (C) the sample transformation is used to transform the non-target samples into transformed non-target samples in order to eliminate discrepancies between the transformed target samples and the transformed non-target samples, so that the non-target base evaluators built from the transformed non-target samples have higher confidence on evaluating the target samples than those non-target base evaluators built without the sample transformation.

11. The method of claim 10 , wherein the sample transformation for the non-target samples is implemented through a kernel embedding method.

12. The method of claim 11 , wherein the kernel embedding method is implemented as solving a transformation matrix optimization problem with optimization goals of: within a common sample space into which the target and non-target samples are transformed, (i) minimizing discrepancies between the transformed target samples and the transformed non-target samples of a same category; and (ii) preserving the volume of knowledge for SS evaluation contained in the transformed non-target and target samples, which is measured by a dimension-category correlation degree of the transformed target and non-target samples.

13. The airspace situation evaluation framework of claim 8 , wherein in step (C) a weight of each of the base evaluators on evaluating a target sample is assigned as the confidence of the base evaluator for evaluating the target sample, then selecting out a portion of the base evaluators with higher weights among all the target and non-target base evaluators.

14. The airspace situation evaluation framework of claim 13 , wherein the weights of the base evaluators are calculated with respect to each of the target samples to be evaluated.